Scalable CAIM discretization on multiple GPUs using concurrent kernels